Questions to Ask Before Hiring an AI Company

Part of our complete guide: How to Use AI in a Small Business.

1. Who owns what you build?

Ask it plainly, and ask what happens to the source code when the work is finished. If the answer is that the system lives on the vendor's platform under the vendor's accounts and cannot leave, you are renting rather than buying, and the monthly figure is the real price rather than the build.

That is not automatically wrong — plenty of good software works that way — but it should be a decision you made knowingly. Our own answer is that you own what we build and the source code transfers to you, which is the position we would advise any owner to look for.

2. Does discovery come before the proposal, or after it?

A fixed quote offered before anyone has looked at your systems is a guess, and guesses get revised once the work starts. The sequence matters: someone should understand what you actually do, which systems hold your information and where the current process breaks before a scope and a number exist.

Vendors who skip that step are pricing a template. It usually surfaces weeks later as a change order, which is the most expensive form of surprise in a small project.

3. What happens in the first month after launch?

Launch is the middle of the work, not the end of it. Real customers ask things no planning session predicted, and the first weeks of transcripts are the most valuable information the project will ever produce. The question is whether acting on that is included or billed as new work.

Ask what support looks like, for how long, and who you call. We include thirty days of post-launch support for exactly this reason, and it is the window where most of the useful tuning happens.

4. What does it do when it does not know the answer?

This is the most revealing question on the list, and the one most likely to be met with a pause. An assistant that invents a plausible answer costs you more trust in one conversation than it saves you in a hundred. The right behaviour is to say it does not know and hand the conversation to a person.

Ask to see that path, not just the happy one. Ask what the handoff is: a transfer, a text to your phone, a logged message someone has to pick up. Our guide to ChatGPT versus a custom AI agent goes into why a general chatbot is far more likely to improvise here than a system built on your own material.

5. Which of my systems does it need to touch?

An assistant that only talks is a brochure. One that books, looks up, updates or sends has to reach your scheduler, your records or your phone system, and that is where the real work and the real risk sit.

Get the list in writing, with read and write marked separately, and find out who on your side has to grant each one. Software vendors can take a while to approve access, and that request is often the longest lead time in the whole project — our guide to how long it takes to build an AI agent covers why.

6. What will it actually be built from?

An assistant is only as good as what it was given. If the answer is generic industry text, you will get generic answers that read fine and are wrong for your business. What you want is your own material: your service descriptions, your policies, your prices, the answers you already give on the phone.

Ask what format they need it in and how messy it is allowed to be. The honest answer is that messy is fine and absent is not, which our guide to training AI on your business documents explains in more detail.

7. Will my customers know they are talking to an assistant?

You want a vendor whose instinct here is disclosure, because the alternative is a customer who works it out mid-conversation and feels tricked. It is also a reasonable thing for a customer to ask, so the assistant should be able to answer it directly.

How it is worded matters less than that it happens early and without apology. Our guide to telling customers they are talking to AI covers the wording that works.

8. How will we know whether it worked?

Agree the measure before the build, not after, and keep it to something you can count: calls answered outside business hours, enquiries that reached a human, appointments booked without a call back. A single number you both agreed on beats a dashboard nobody reads.

Be wary of a vendor who will not commit to one. If nobody defines success at the start, every conversation about whether it is working becomes a matter of opinion, and the opinion that counts will not be yours.

9. What does it cost to walk away?

Ask the uncomfortable one last. If you ended the relationship tomorrow, what do you keep? The code, the content you supplied, the conversation history, the phone number, the accounts the whole thing is hosted under? Who is the registrant on each one?

The answer tells you how much leverage you are handing over. We work without lock-in because we think that is the only version a small business should accept, and it is worth insisting on wherever you buy.

Two Things That Are Not Questions

Watch what gets volunteered. A company that tells you what AI will not do for your business, unprompted, is more useful than one that says yes to everything, because the second kind has not thought about your business at all. Our guide to why custom AI agents beat off-the-shelf tools is largely a list of those limits.

And watch the first version they propose. Someone suggesting a narrow build that handles one job well is thinking about your result; someone proposing to automate everything at once is thinking about the invoice. Narrow reaches customers sooner and teaches you more.

About Lexington AI Solutions: We are a veteran-owned AI automation consultancy based in Lexington, KY, serving businesses throughout the Bluegrass region. Learn more about our team.